ECONOMY BOX

'26/11/19 UPDATE

PRICES AND INFLATION / FIELD NOTES

In October, consumers expected 2.7% inflation over the next year. Is that the average answer or the middle one?

The 2.7% expected inflation in the Bank of Korea's October consumer survey is the median of answers given in ranges. Recompute it from the release's table of answers, see why an average built on assumed range midpoints comes out higher (about 2.9% under this note's assumptions), and set perceived inflation of 2.9% next to actual consumer price inflation.

In its Consumer Survey results for October 2026 (in Korean), released on 27 October 2026, the Bank of Korea put consumers' expected inflation over the next 12 months at 2.7%. In the same table, perceived inflation, the answer to how much prices are felt to have risen over the past year, was 2.9%, and expected inflation 3 years and 5 years ahead (the expected rise over the third and the fifth year from now) was 2.6% each. A note under the table says all four figures are medians. They are not averages, the sum of the answers divided by their number, but the value in the middle when the answers are lined up.

After reading this, you will be able to recompute the 2.7% yourself from the table of answers in the release, see why the same answers counted as an average would give a higher number, and see what appears when perceived inflation of 2.9% is set next to actual consumer price inflation.

Nobody wrote down 2.7%

The Consumer Survey sends the same questionnaire every month to 2,500 urban households across Korea. In October it was fielded from 13 to 20 October, and 2,238 households answered. The price questions have no box for a number. In the latest questionnaire the Bank of Korea has published, the October 2024 revision, the question on how much consumer prices will rise or fall over the next year offers nine choices: a fall, six ranges one percentage point wide from "0% to under 1%" up to "5% to under 6%", "6% or more" and "don't know". The questions on the past year and on 3 and 5 years ahead use the same nine choices.

The statistical description in the Bank of Korea's ECOS database says the same: expected and perceived inflation are surveyed in ranges 1.0 percentage point wide and calculated as medians. Its record of changes also notes that in January 2013 the calculation of expected inflation switched from a weighted average to the median. So 2.7% is neither a number someone wrote down nor an average of the answers. It expresses where, inside the 2-3% range, the middle answer falls when the chosen ranges are lined up from lowest to highest.

The Bank of Korea's glossary of 800 economic and financial terms (2026, in Korean) counts expected inflation among the key indicators a central bank aiming at price stability watches. Its explanation is that economic agents' subjective view of future inflation feeds into wage bargaining, price setting and investment decisions, and so in the end into actual inflation.

Recomputing 2.7% from the table of answers

Page 4 of the release attachment lists the share of answers in each range for the next 12 months. In October they were: fall 0.6%, 0-1% 7.4%, 1-2% 19.4%, 2-3% 31.0%, 3-4% 20.7%, 4-5% 7.4%, 5-6% 4.6%, 6% or more 5.3%, and don't know 3.5%. The range chosen most often is 2-3%, but that alone does not give 2.7%.

Shares of answers on consumer prices over the next 12 months, October 2026: fall 0.6, 0-1% 7.4, 1-2% 19.4, 2-3% 31.0, 3-4% 20.7, 4-5% 7.4, 5-6% 4.6, 6% or more 5.3; don't know, 3.5, is left out. The running total is 27.4% up to 2% and 58.4% up to 3%, so half of the 96.4% who gave a range, 48.2%, falls in the 2-3% range. The line chart below shows, from February to October 2026, perceived inflation at 2.9, 2.9, 2.9, 3.0, 3.0, 3.0, 3.0, 3.0 and 2.9; expected inflation for the next 12 months at 2.6, 2.7, 2.9, 2.8, 2.8, 2.7, 2.7, 2.7 and 2.7; and actual year-on-year consumer price inflation at 2.0, 2.2, 2.6, 3.1, 3.2, 2.8, 3.1, 2.9 and 2.8.
Figure 1. October's distribution of answers for the next 12 months (top), and perceived inflation, expected inflation and actual consumer price inflation from February to October (bottom). Drawn from the tables on p. 4 of the Bank of Korea's October 2026 Consumer Survey release attachment and the ECOS 901Y009 index.

The calculation takes three steps. First, leave out "don't know". The ranges with a number add up to 96.4%, and half of that is 48.2%. Next, add the shares from the lowest range up. Up to the 1-2% range the running total is 27.4%; up to the 2-3% range it is 58.4%, so the middle answer lies inside 2-3%. Finally, assume the answers are spread evenly inside that range and split it on a straight line. The middle is 20.8% of answers (48.2 − 27.4) into the 31.0% in the 2-3% range, so 2 + 20.8 ÷ 31.0 = 2.671, which rounds to 2.7% at one decimal place.

Step 1: the eight ranges from fall to 6% or more add up to 96.4; don't know, 3.5, is left out. Half is 48.2. Step 2: the running total is 0.6 plus 7.4 plus 19.4, or 27.4, up to 2%, and 58.4 up to 3%, so 48.2 lies in the 2-3% range. Step 3: 2 plus (48.2 minus 27.4) divided by 31.0 is 2.671, or 2.7% to one decimal place. Step 4: the same method gives 2.706, or 2.7, for September's next 12 months; 2.618 and 2.610, both 2.6, for October 3 and 5 years ahead; and 2.636 and 2.620, both 2.6, for September. Adding don't know only to the total, as if those answers sat above every range, would give 2.763, or 2.8, for September, which does not match the published figure.
Figure 2. The median recomputed from the published shares alone. The same method matches all six published figures for September and October: next 12 months, 3 years ahead and 5 years ahead.

Steps two and three are the formula the Bank of Korea's handbook Understanding the Consumer Survey (January 2014, in Korean) gives on page 41 for calculating expected inflation: answers inside a range are assumed to be spread evenly, and the value of the middle answer is estimated by linear interpolation. What the handbook does not cover is how the first step treats "don't know": it describes the total to be halved only as the people surveyed.

So apply the same calculation to the other distributions in the release. September's next-12-month answers give 2.706; October's 3-year and 5-year answers give 2.618 and 2.610; September's give 2.636 and 2.620. Rounded, that is 2.7, 2.6, 2.6, 2.6 and 2.6, so together with October's next 12 months all six published figures match. If "don't know" is added only to the total instead, which amounts to counting those answers above the highest range, September's next 12 months becomes 2.763, or 2.8%, and five of the six no longer match.

Still, the handbook dates from 2014, and leaving out "don't know" is an assumption that rests on these matches rather than on a document. According to the ECOS description, regional and age weights are applied to the survey results; here the shares are used exactly as published.

Averaged at range midpoints, it comes to about 2.9%

To count the same distribution as an average, each range needs one representative value. Closed ranges can use their midpoints (0.5 for 0-1%, 2.5 for 2-3%), but "fall" and "6% or more" are open-ended, so a value has to be assumed. Here a fall counts as −0.5 and "6% or more" as 6.5. Multiplying each value by its share and adding gives 275.5; divided by 96.4 that is 2.858, or 2.9% rounded. It is a worked example with assumptions, not an official statistic.

Assumption: each range counts at its midpoint, 0-1% as 0.5, 1-2% as 1.5 and so on; the open-ended ranges count as −0.5 for fall and 6.5 for 6% or more. Weighted average: −0.5×0.6 + 0.5×7.4 + 1.5×19.4 + 2.5×31.0 + 3.5×20.7 + 4.5×7.4 + 5.5×4.6 + 6.5×5.3 = 275.5; divided by 96.4 this is 2.858, or 2.9%. Answers of 4% or more make up 17.3%, answers below 1% including fall 8.0%. Counting 6% or more as 8 gives an average of 2.940, as 10 gives 3.050, while the median stays at 2.671, or 2.7, whatever value the open-ended ranges get.
Figure 3. Averaged at assumed range midpoints, the same distribution gives about 2.9%. The average moves with the value chosen for "6% or more"; the median does not.

The average sits about 0.2 percentage points above the median because of the shape of the distribution. Answers of 4% or more make up 17.3%, more than twice the 8.0% below 1%, fall included. A long upper tail pulls the average up, while the median only looks at where the middle answer falls. Counting "6% or more" as 8 instead of 6.5 raises the average to 2.940, and as 10 to 3.050, yet the median stays at 2.671 whatever value the open-ended ranges are given. As long as the median falls in a closed range, it needs no value for the open-ended ranges; that is a property of the median in a survey that asks in ranges.

Comparing September with October shows why the distribution is worth reading. In October, the top range, "6% or more", rose from 4.6% to 5.3% and "0-1%" near the bottom from 6.3% to 7.4% ("5-6%" fell, so 5% or more stayed at 9.9% in both months), while "2-3%" fell from 32.0% to 31.0%. By the formula above, the median moved only from 2.706 to 2.671, and both months were published as 2.7%. The spread of answers can change while the published figure stays the same.

Perceived inflation of 2.9% and actual prices

Perceived inflation of 2.9%, how much prices are felt to have risen over the past year, is also the median of a range question. The release does not print its distribution, but the ECOS table its appendix points to, 6.2.3 Consumer Tendency Survey (Expected Inflation) (National), does. In October the answers were: fall 0.1%, 0-1% 5.9%, 1-2% 16.1%, 2-3% 29.0%, 3-4% 22.2%, 4-5% 9.3%, 5-6% 7.2%, 6% or more 7.5%, and don't know 2.7%. Counted the same way, half of the 97.3% who gave a range, 48.65%, falls in the 2-3% range: 2 + (48.65 − 22.1) ÷ 29.0 = 2.916, or 2.9%, the published figure. Widened with the same table to the nine months from February to October, the method matches all 18 published figures: nine for perceived inflation and nine for expected inflation over the next 12 months. That supports the assumption that "don't know" is left out, but it is not confirmation from a document.

To set it next to actual consumer prices, take the all-items consumer price index in ECOS (901Y009, 2020 = 100) and compute the rise over the same month a year earlier. The latest figure available while the survey was in the field, 13 to 20 October, was September's, released on 2 October under the official schedule: (120.43 ÷ 117.06 − 1) × 100 = 2.879%, or 2.9%. October's rate came out on 3 November, after the survey had closed, in the Ministry of Data and Statistics' Consumer Price Trends, October 2026 (in Korean): (120.73 ÷ 117.42 − 1) × 100 = 2.819%, that is 2.8%. The Ministry plans to rebase the index, including its items and weights, to 2025 and publish it in December 2026, after which figures published on the 2020 base may change.

The lower panel of Figure 1 shows that from February to October perceived inflation stayed at 2.9-3.0%, while actual inflation moved between 2.0% in February and 3.2% in June. Expected inflation for the next 12 months was 2.6-2.9%. Perceived inflation is the median of survey ranges, and consumer price inflation is the change in an index built from the prices of many items, so the two numbers are made differently from the start. The comparison is useful for seeing the size of the gap in the same month, not for deciding which one is right. On the actual-price side, what each of October's two core inflation measures leaves out is covered in Both of Korea's core inflation measures were 2.8% in October. What does each one leave out?, and why the same survey's consumer sentiment index is read against 100 is covered in Four of its six components are below 100. Why is consumer sentiment 106.6?.

Reading the distribution behind the headline figure

When expected inflation is published, open the table of answers next to it. With the range that holds half of the shares without "don't know", the running total below it and that range's own share, you can apply the handbook's formula yourself; from February to October it gave the published figures. For perceived inflation, the distribution is in the ECOS 6.2.3 table. The same formula also lets you gauge how close the unrounded value is to a rounding boundary: October's next-12-month value of 2.671 was 0.021 above 2.65, the point at which it would round down to 2.6.

The figures here are the October survey results and October consumer prices, checked on 9 November 2026. The Bank of Korea's release calendar has the November survey results on 24 November, and the Ministry of Data and Statistics' schedule has November consumer prices on 2 December.

THE END
RSS feedBack to the library
COMMENTS BOX

Add your perspective.

Share a question, another approach, or something you have tried.

Newest first

Checking sign-in…

Loading comments…